In khodosevichlab/CellAnnotatoR: Automated marker-based cell type annotation

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This vignette shows annotation of the first BM dataset from the Conos tutorial

library(CellAnnotatoR)
library(pagoda2)
library(dplyr)
library(ggplot2)
library(pbapply)
library(magrittr)

theme_set(theme_bw())

Pre-processing

Let's load and pre-process data:

cm <- system.file("extdata", "panel.rds", package="conos") %>% readRDS() %>% .[[1]]

p2 <- basicP2proc(cm, n.cores=4, min.cells.per.gene=0, n.odgenes=2e3, 
                  get.largevis=FALSE, make.geneknn=FALSE, get.tsne=T)
p2$getKnnClusters(type="PCA", method=conos::leiden.community, n.iterations=10, 
                  resolution=6, name="leiden")

Here we use such large resulution because it will be used for the annotation refinement: for high-resolution annotation, clustering resolution must be high as well.

clusters <- p2$clusters$PCA$leiden
conos::embeddingPlot(p2$embeddings$PCA$tSNE, groups=clusters, shuffle.colors=T)

Prepare data for annotation:

marker_path <- system.file("extdata", "bm_cb.md", package="CellAnnotatoR")
clf_data <- getClassificationData(cm, marker_path)

Annotation

Now we can run annotation:

ann_by_level <- assignCellsByScores(p2$graphs$PCA, clf.data=clf_data)
cowplot::plot_grid(
  conos::embeddingPlot(p2$embeddings$PCA$tSNE, groups=ann_by_level$annotation$l1),
  conos::embeddingPlot(p2$embeddings$PCA$tSNE, groups=ann_by_level$annotation.filt$l1),
  nrow=1
)

To improve quality we can also pass clusters to the annotation procedure:

ann_by_level <- assignCellsByScores(p2$graphs$PCA, clf.data=clf_data, clusters=clusters)
cowplot::plot_grid(
  conos::embeddingPlot(p2$embeddings$PCA$tSNE, groups=ann_by_level$annotation$l1, shuffle.colors=T),
  conos::embeddingPlot(p2$embeddings$PCA$tSNE, groups=ann_by_level$annotation.filt$l1, shuffle.colors=T),
  nrow=1
)

Finally, to validate results we can plot the markers:

plotSubtypeMarkers(p2$embeddings$PCA$tSNE, p2$counts, clf.data=clf_data, n.col=3, size=0.2, build.panel=T)

khodosevichlab/CellAnnotatoR documentation built on June 29, 2022, 9:12 p.m.